AI Tutor, Human Tutor or Videos: An Honest Comparison

Once you have decided your child needs math help beyond the classroom, the next question is which kind. A private tutor at $50 an hour. A video library, free. An AI tutor for the price of two coffees a month. They are not variations on one product; they do genuinely different things, and the research on them differs more than the marketing suggests.

We build an AI math tutor, so read this knowing where it comes from. We have tried to write the comparison we would want if we were choosing, including the cases where our own category is the wrong answer.

What the research actually says#

Start with the famous number, because it is usually quoted wrongly.

Benjamin Bloom (1984) reported that students tutored one-to-one performed about two standard deviations above students in a conventional classroom — the "2 sigma problem." It is the founding claim of the whole tutoring industry.

It has not held up at that magnitude. Kurt VanLehn's 2011 review, which compared human tutoring against computer-based tutoring systems across a large body of studies, found human tutoring producing effects of roughly 0.8 standard deviations rather than 2.0. Still a large effect. Not the mythic one.

The more interesting half of VanLehn's finding is the comparison. Step-based computer tutoring systems — the kind that respond to each step a student takes rather than only to a final answer — came out in a similar range to human tutors, around 0.76 against 0.79. Kulik and Fletcher's 2016 meta-analytic review reached a compatible conclusion, with a median effect around 0.66 standard deviations.

Two things follow. First, the gap between a good human tutor and a well-designed tutoring system is smaller than intuition suggests. Second — and this is the part that matters for choosing — what made both effective was the same mechanism: responding to the individual student's reasoning, step by step, in the moment.

That mechanism is the thing to shop for. It is also exactly what video lacks.

Videos#

What they are good at. Free, unlimited, available at 10pm, and often genuinely well explained. A good video series is the cheapest way to see a method demonstrated clearly. For a parent who needs to understand how long division is being taught now before helping, a five-minute video is the fastest route there is.

Where they fall down. A video cannot find out what your child is thinking. It delivers the same explanation, in the same order, to a child who is confused about place value and a child who is confused about notation. If the explanation happens to match the confusion, it works. If it does not, the child watches to the end and remains stuck, often now believing that they watched an explanation and still did not get it — which is corrosive.

Chi and Wylie's ICAP framework (2014) puts this in order: students who generate and construct ideas learn most, those who actively work with material learn less, and those who passively receive it learn least. Watching is the passive mode. That is not a criticism of any particular channel; it is a property of the format.

Verdict. Excellent reference material. A poor primary intervention. Use them to look something up, not to close a gap.

Human tutors#

What they are good at. A skilled human tutor is still the best thing money can buy for a child in trouble. They read frustration on a face and change course. They notice that the real problem is a fourth-grade idea, not this week's chapter. They build a relationship, and a child who does not want to disappoint their tutor will try harder than one who does not want to disappoint an app. For a child with a diagnosed learning difference, severe math anxiety, or a gap spanning multiple grades, this is the right call and nothing here should talk you out of it.

Where they fall down. Cost and frequency, which are the same problem. At around $50 an hour — an illustrative rate; Texas publishes no official average — a weekly session over a 36-week school year runs to roughly:

$\$50 \times 36 = \$1{,}800.$

For that, your child gets math help once a week. But gaps close through frequent short practice, not weekly long sessions, and the six days between sessions are exactly when homework happens. Most families cannot sustain three sessions a week for the months a real gap takes, so the frequency that actually matters is the thing the budget cuts first.

Quality also varies more than the price does. "Good at math" and "good at teaching math to an eleven-year-old who has stopped believing they can do it" are different skills.

Verdict. The strongest option when the problem is serious or emotional, and the hardest one to afford at the frequency that works.

AI tutors#

What they are good at. Availability and cost, which between them buy frequency. A subscription in the region of $30 a month against $200 a month for weekly human sessions changes what a family can sustain. And it is there at 9pm on a Wednesday when the homework is actually happening, which is when help is most needed and least available.

A well-designed one also does the thing the research identified: it responds to the step your child just took, adapts the next question to what they actually said, and does not move on because the lesson plan says so. It never gets impatient at the fourth repetition, which matters more than it sounds for a child who has learned to expect adult frustration.

Where they fall down. They do not read a face. A child who is quietly falling apart may keep typing, and the system sees answers rather than distress. They cannot call the teacher, and they cannot notice that the real problem is that your child did not sleep. For a diagnosed learning difference, a human professional should lead and software should support the practice in between.

And the category is full of things that are not tutors. Many products described as AI tutors are answer engines: photograph the problem, receive the solution. Those make homework faster and learning slower, and a child using one can produce a term of perfect homework alongside falling test scores.

Verdict. The best value for daily practice and homework support, provided you check that it asks rather than tells.

The test that separates them#

Whatever you are considering, run this before paying for anything. Give it a problem from your child's grade and type "I don't know" twice.

• An answer engine gives you the full worked solution. Walk away.
• A tutor asks something smaller. For "what is $\frac{2}{3}$ of 24?", it should ask what one third of 24 is, wait, and then ask what two of those would be.

That single test is worth more than any feature list. Our longer write-up of what to look for is in how AI tutoring helps struggling math students, and the teaching approach underneath it is in the Socratic method in math.

A simple way to choose#

Your child is broadly fine and needs homework support. An AI tutor or a good practice app. Frequency is what you are buying.

Your child has a specific gap you can name. Targeted daily practice on that gap, from any source that adapts. Fifteen to twenty minutes, four or five days a week, closes most single-grade gaps in eight to twelve weeks.

Your child is two or more grades behind, or something is emotionally wrong. A human tutor or an educational psychologist first. Use software for the practice in between, not instead.

You do not know which of these you are in. Find out before you spend anything. This is the most common and most expensive mistake — a tutor working at grade level cannot fix a problem that started two grades earlier, and you will pay for months before discovering it. The five signs a child is falling behind describes the pattern, and the free Math Check-Up gives a ten-question snapshot for your child's grade in about ten minutes, with no account and no payment.

Money is genuinely tight. Videos plus your own time, using the questions in why productive struggle works, beats nothing by a wide margin. The most valuable input is an adult who asks "how do you know?" and waits, and that costs nothing at all.

What none of them do#

No tutor of any kind can supply the thing that determines the outcome: consistency. Fifteen minutes, most days, for a few months. Every option above works if that happens and none of them work if it does not.

If you remember one thing from this comparison, make it that. Choose the option your family will actually keep doing on a Wednesday in February, and the rest matters considerably less than the marketing implies.